We tested the predictions of three models (female preference; hotspot; predator avoidance) on lek formation in the fallow deer population of San Rossore, Tuscany. We collected behavioural observations in two leks and radiotracking data on 67 deer over 7 years. Two deer sub-populations were present in the northern and southern sides of the area, respectively, the two sectors being delimited by a river and including one lek each. Predictions were tested for one lek (SG), located in the south-side where we set up our 7-year radiotracking program. Data from a second lek (FO, north-side) were used to test those predictions which imply the occurrence of multiple leks in the same population. We showed that the majority of females made one single visit to one lek, only during the rut. The lek was located outside areas of higher female traffic and home range overlap, and females increased home range sizes during the rut to reach it. Twilight routes of females never crossed the lek; instead, females walked atypical routes and at a faster pace to reach the lek and mate. The distance between the two leks was higher than the average diameter of female home ranges, and only one lek was present within female home ranges. Males reached the lek one month before the arrival of females, corroborating that lekking is a female-initiated process (females moving towards large clumped male aggregations) rather than a male-initiated process (males moving towards female hotspots). Our results supported the female preference model, and rejected the predictions of the hotspot model. Also, leks were located far from areas with higher predation risk, supporting the predator avoidance model. The position of lek SG resulted ‘handy’ at the sub-population level because of the optimal trade-off between travel costs for females to reach it and avoidance of human predators.